DS102 - MNIST

A short data science project that performs inferences on handwritten digits with a Neural Network Model trained on the MNIST Dataset using PyTorch.

Fulfilled as part of the requirements for DS102

To Use

Drawing Inference Web App (Online)

No Installation Needed! Simply visit MNIST Draw - KUNOSPSIM to access the drawing inference Web App.

Drawing Inference Application (Local)

Note: Requires pyxel and pytorch to run

Simply run pip install pyxel pytorch to install.

draw.py - A python script written with pyxel, that launches a drawing app that performs inferences whenever a user draws a digit.

To run simply run pyxel run draw.py in the root directory

Controls

• LMB Draw a white pixel with a radius of 2
• RMB Erase a white pixel with a radius of 2
• Q Clear the screen

Training Notebook

DS102_KANUPSIM_MNIST.ipynb contains the code for training the model through pytorch. Simply run each cell to generate a new model. (The last cell runs a 'quick' test on the test dataset to test accuracy)

Attributions

Training

PyTorch Docs were highly utilized to make the notebook for training the NN. Most functions for training and testing the model from the dataset were from the quickstart guide.

Web Application

Streamlit was used to create, host, and deploy the Web Application, the Drawing 'Canvas' component was made with andfanilo/streamlit-drawable-canvas.